A Cognitively Grounded Bayesian Framework for Misinformation Susceptibility

TL;DR

This paper introduces BPL, a cognitively grounded Bayesian model extending RSA theory with bounds reflecting human cognitive limits, to predict misinformation susceptibility and related phenomena.

Contribution

It develops a novel Bayesian framework incorporating cognitive bounds to better model susceptibility to misinformation and validates it on benchmark datasets.

Findings

BPL achieves competitive veracity classification performance.

Supports the depth-mismatch paradox in misinformation susceptibility.

Provides insights into annotator disagreement and vulnerability to different information types.

Abstract

In this (work in progress) paper, we present Bounded Pragmatic Listener (or BPL), a cognitively grounded Bayesian framework for modelling susceptibility to information disorder. BPL extends Rational Speech Act theory with three cognitively motivated bounds derived from the bounded rationality literature with a) a recursion depth bound (that emphasises working memory limits);b) a prior compression parameter (which is oriented at capturing information bottleneck); and c) an availability sample size (that operationalises importance sampling with saliency-weighted proposals). This allows us to test predictions about misinformation susceptibility, annotator disagreement, and the differential vulnerability to mis-, dis-, and mal-information as defined in the Information Disorder framework. We validate BPL on the LIAR and MultiFC benchmarks showcasing competitive veracity classification and experimental support for the depth-mismatch paradox.